Experimental Design & Observational Studies
From slides: Longitudinal
- One group of subjects is followed over time -Subjects are compared to themselves at different points in time -Subjects must be measured/interviewed more than once.
Diagnosis
- so if a doctor knows that this child has been vaccinated, or not been vaccinated, they could potentially be influenced as to whether or not they think this is polio, especially in those mild cases. So what they wanted to do with the placebo control was to make sure that no one involved at the actual personal level knew who had been vaccinated and who had not. So you'd have two children in the same family and one of them maybe got the vaccine and the other didn't and they both ot sick, the doctor doesn't know which one so he will be more impartial on whether his diagnosis is polio or not.
No consent group
-Polio rate was smaller -Generally from lower income strata -more natural (early) immunity to polio
Cross-sectional
-Subjects are only measured once. -Subjects are compared to each other, or to subjects from another study -Do not use to draw conclusions about what happens as individuals age, etc.
Consent group
-higher income -less natural (early) immunity to polio -higher incidence of polio
Common Confounding Factors
1. Age 2. Gender 3. Socio-economic status 4. Education 5. Sexual history 6. Genetics 7. Occupation 8. Religion 9. etc.
Controlled experiment
1. Researchers decide which subjects are in treatment and control groups. 2. researchers impose some change in behavior on the participants 3. Confounding factors are controlled for using randomization 4. IF DONE WELL AND ARE REPLICATED, CAN ALLOW US TO (CAUTIOUSLY) INFER CAUSATION.
Observational studies
1. Subjects assign themselves to the different groups, usually by their now choices or behavior 2. Establish associations between variable 3. Confounding factors are controlled for after data is collected 4. DO NOT IMPLY CAUSATION UNLESS MULTIPLE STUDIES OVER A LONG PERIO DOF TIME ALL GIVE THE SAME RESULTS. (EVEN THEN, BE CAREFUL).
How do you determine if a drug is effective?
1. need to give it to a lot of people 2. Need to be able too compare to a group not getting the drug
• So it's also a social responsibility that if you are healthy enough to have the vaccine that you do that to protect others who can't. No medicine is perfect, no vaccine is going to be
100% effective
No vaccine is going to be
100% safe. There will be accidents and situations where there are problems. If it happens to you its tragic, but just like car accidents are tragic and plane crashes, etc.
. Not only do we use control different than the general world, but we use it
3 different ways in statistics, so this is a really problematic word
But once we've doen that we can pretty confidently say that
A causes B, which is what happened when we developed the Polio vaccine. That's the kind of process we went through.
You get a whole group of people, you randomly divide them in half, you give half of them A and half of them B. Nobody knows which half gets which one except for some researcher back with the Excel Spreadsheet who never sees the patients. So somebody knows who is getting which one, but not anybody who's directly involved with it. And if there's a difference between treatment and control, then we can with some confidence say that
A causes B. Not with complete confidence, we need to be able to replicate that again
What allowed Tommy Francis to say that the polio vaccine worked?
A controlled experiment that was double blind using placebo controls
Other types of controls- There are other types of controls. There are sometimes when it's not possible to do a double-blind controlled experiment
There are somethings where you can't do a placebo, like surgery. It's very difficult to get people to sign up for placebo surgery. • I read a study where they did a blind experiment where they were testing arthroscopic surgery for knee injuries. And they had one group of people where they went in and cut the knee open and sewed it up, and then another group where they actually did the procedure, so that was a blind study where the person didn't know, but obviously the surgeon can't be blind to that kind of a scenario.
Maybe not having cancer made people more religious. That's as plausible as the other way around, if all you have is an observational study and this established association,
There's no knowing which way that cause goes.
Confounding factors
When the treatment group and the control group differ in some respect other than the treatment,
You can tell from the numbers that the vaccination rate is about the same, or the rate of diseases in the vaccine groups are about the same, 28 per 100,000. But we had
a much larger difference between the control group and the treatment group under the placebo controls.
• We can narrow it down by how the word is used. If we are using it as a noun, so we're talking about a control, it is
a person or a subjecti n the study who does not get the treatment.
• So right there we are seeing a big difference because we've got a lot of people, a treatment control group and clearly there's a difference in the rate of polio between those two groups. • When Tommy Francis came on board he said no we can't do it this way we've got to have a placebo control group. And this is really late in the game they want to get this trial done for the summer of 1954 because polio was a season disease and it started early in March or April in the South where it was getting warm and then it would spread up into Canada and most cases were done by November. So they really need to get moving because they're having talks about it in January. And they get this big meeting that they had they said no we have to have
a placebo group
This kind of a design aslo lets you use
a smaller sample size in your study, fewer people and still get good results from the experiment
. So half received the
actual live vaccine, and then half got an empty shot of saline solution
A controlled experiment
adj. is research where investigators determine who will be in the control group and the treatment group.
A confounding factor is something that
affects Variable A and also affects Variable B.
Because subjects are NOT assigned to treatment and control using
an objective chance procedure, differences can be caused by SOMETHING OTHER THAN THE TREATMENT
So if someone's coming in and saying you guys do this you guys do that it's going to be an experiment. If nobody did that, it's going to be
an observational study
o Now as a medical person, it's a little bit hard to conclude that something in religion is or is not causing cancer. We don't tend to think that our belief system, what's going on in our brain actually causes cancer, but it's not beyond the realm of possibility. Is there something about going to church or believing in a particular philosophy that somehow protects against cancer. So again the association exists. There's no doubt that there's an
association, but we don't necessariy know that one is causing the other.
The reason that it's important is because when we go to construct treatment and control groups we want those two groups to
be as similar as possible. We want the only difference to be did they get the treatment or not? And if we randomize we end up with about half in both groups.
**Controlled experiments can lead to
causal inference. You can infer that A causes B.
Historical controls- historical controls have to do with
comparing people who receive a treatment today with people who received a different treatment in the past,
We call those differences
confounding factors
So let's talk about some of the issues that caused the differences between the placebo controlled and the observed controlled studies. (see slides) The first big one was
consent. Another issue is diagnosis. Lastly the hugest difference comes from the placebo effect
A problem we might have with historical controls compared to this called
contemporaneous controls
What do you do with confounding factors? Here we have to introduce another word that has lexical ambiguity-
control
Okay, by getting a homogenous group of people and randomly assigning them to treatment and control, that's how we
control for confounding factors in an experiment.
And then we
control for confounding factors. So we are going to use the word again to talk about what we do with confounding factors to try to figure out if we can demonstrate a causal relationship, even though this is just an observational study.
The biggest determination is
did the researchers tell people what to do? Or did they just gather this information on stuff that people had already done?
And we can end up with some pretty significant
differences between the treatment and control besides the treatment.
They are compared to
each other. So I go out and take a big group of people and I compare people within that group to each other
For observational studies, especially if you don't have an idea what's going on, you
generally need several thousand people at the least to get enough subgroups to have information, enough people in the sub groups is basically what's going on with observational studies
But the placebo effect is a big problem, in fact they're starting now to notice twhat they're calling the nocebo effect. That if you
give someone a medicine and tell them all the side effects they will go home and manufacture the side effects, where if you don't tell them the side effects theya re fine
One of the things that hopefully you get from this class, is the abiity to
go and look at the original research study. And you can to the website where the article is and now they'll even have tab that indicates the randomization method. They'll talk about their methodology and who how it ws decided who received what.
So again anytime somebody tries to hand select is almost always
going to end up with some sort of bias. We just can't make purely random decisions.
longitudinal studies are a little better at what?
helping us establish those causal factors again
o It's not always easy to figure it out,
hence research.
Generally pretty safe assuming that it was double blind, because
in the medical fields everyone knows if you can do double blind that you should do double blind.
From slides: A control
is a subject in a study who did not get treatment
When we're worried about controls and treatment, the problem that we're concerned with is,
is our treatment group the same as our control group.
So if we are using this older groups as the controls, compared to this new group with the treatments, what we're concerned about is,
is there some difference between the two groups besides the treatment that they were given?
Those people can't be vaccinated so by having everyone else in the society vaccinated,
it reduces the chance those people who can't be vaccinated will come in contact with the disease.
The no consent group was
less likely to get the disease in the 1st place so there would be bias against the vaccine. (Meaning the no consent group may be less likely to get the disease so the vaccine won't show as much as an affect because it will work much better on the consent group)
What happens with an observational study is we say,
look, people who have this condition, we'll call it condition A, also tend to have condition B. So people who smoke tend to get lung cancer, and we start to notice this as time goes on. So we have condition A and Condition B and there seems to be some kind of a relationship between them
♣ So the idea with a cross-sectional study is you would compare old people to young people. And old people have less bone density than young people, Does that mean bone density decreases as you age? Maybe, or it could just be that something happened in this older generation that caused their bone density to go down as opposed to the younger generation. So in order to determine that bone density actually decreases as we age, you would have to
measure somebody as a young person and then measure that same person again later as an old person and see if their bone density decreased.
We might have different
measurements that we're taking historically compared to the measurements we are taking now
In an observational studies we can't do that, so we control for it afterwords by
messing around with the data
• Part of the reason we're so afraid to fly even though flying is much safer than driving in an accidents per mile way, I think it's because we don't feel like we have any control. • When you're driving you feel you have some control, but in an airplane somebody else is driving and we may not understand how it's flying. So like last you're much more likely to get into a fatal car crash on the way to the doctor to get a vaccine than you are to be hurt by a vaccine, but we feel like we're in control when we're driving our car to the doctor so our level of fear about that thing is
much moer muted. Also we drive around and are more used to the car, we see accidents all the time, so we're a little immune to the fear of driving a car. Even though it's dangerous.
That's the basic idea of what happens. Back in actuality what happens is we have very complicated mathematical models that are run on computers that helps us determine these kidns of things. It's called in case you ever want to know,
multiple linear regression. (not going to do this in this class).
Use of randomization-did it talk about randomization?
o It didn't necessarily say. It just said half were told to repeat their favorite curse words and the others didn't. So in this case the randomizations may not be quite so important. But because maybe order makes a difference you would still want to randomly decide who did the swearing first and who did the non-swearing first, again to eliminate any kind of bias that you could get that way.
o So in this country it seems that wealthier mothers tend to breastfeed more than les wealthy mothers. That might not be true in other countries, but it might be flipped. But it probably does have some kind of effect, so wealth and job status affect how mothers chose to feed their children. o So we can say that without having to know whetehr it goes one way or the other, we can come up with reasons why that might be the case. Now how does your other question is, how does socioeconomic status possible affect vision? In this case we aren't necessarily talking about corrected vision, this was just measuring their ability to see basically in 3D.
o One of the nice things about socio-economic status as a confounding factor is that it affects so many other things. It specifically affects diet and nutrition and access to healthcare. o Since we are talking about something that involves the body and how it works, access to nutrition and healthcare is going to, we don't know all the ways that it affects different parts of our life, and our bodies and how they function. We haven't quite figured out the optiml way of doing everything, and even once we did, people are variable that it would, probably be good for this group but not that group. o So there are definitely some effects going on there. So you can just at this point say nutrition and healthcare probably have or could have something to do with vision. That's what makes this a confounding factor. Socioeconomic status affects the choice to breastfeed for the reasons we talked about and it also can affect things like nutrition which could affect vision.
• Vaccine safety and effectiveness
o There is always people and there will always be people who see conspiracy where there is none. Right now we are dealing with a lot of outbreaks of preventable diseases because people have decided that vaccines are optional or bad or something like that.
• Student question: How do they find people to participate in the study? A lot they will call for volunteers and then screen out people who don't have disease. Say we want people over the age of 30 who have moderate to severe acne.
o They will get volunteers, which does actually bias the outcomes of the study. Because it may be that people who volunteer for a study have something different about them compared to the regular population. And it may be that only volunteers are going to get the benefits of the drugs. But again, we're limited by what we are able to do. We don't live in a totalitarian society where we can just go out and say you're going to be in this drug study. They have to volunteer.
o Student answer- Doesn't specify blind or double blind- Teacher answer-
o Well if there was a placebo and it was well disguised, which the last paragraph talked about that means it was conducted blind. But you're right that it doesn't talk about whether or not it was double blind.
Is this blind?
o You couldn't do something like that blind either right? Obviously the woman will know if she is breastfeeding or formula feeding her baby.
It's in bold at the bottom, ***In a controlled experiment, if it's done well, so all the things we've talked about
o control, randomized, double blinds, all those things, if it's done well and it can be replicated and has been replicated. Then you can cautiously start to think that maybe factor A causes factor B.
But again, you have to be able to explain
o how it affects both condition A (religion affects the rate of circumcision) and condition B (religion affects sexual behavior which affects the low cervical cancer rates in women. ♣ So it can get pretty confusing.
***Observational studies you can't do that. They do not
o not imply that A causes B. It's just as equally likely that B causes A in an observational study. It can go either direction, that's why there's two arrows on the diagram.
So if we do it among college students we need to then go out and try it among the elderly or children, or people on the other side fo the world who eat different food, we need to be able to
o replicate that over and over.
There are a lot of the chemicals that we take in as medicines that our brain actually manufactures themselves, specifically
o specifically pain response, and endorhpins, and immune response, our brain makes those chemicals as well. Our bodies create white blood cells and endorphins and things that relieve pain.
So the subjects themselves, when we went to investigate what's causing lung cancer, could it be smoking? All we could do was
o was look at people who had smoked and compare them to people who hadn't smoked. o And the subject's decisions themselves decided whether they were in the treatment group or the control group.
What about randomization for this study? o Student answer- I put there was because it didn't specify who got it. Teacher answer- Again
o you can probably safely assume from this kind of thing that it was randomly assigned because that's what you're supposed to do. o If you're concerned again go and look at the original article.
When I ask you to describe or explain why something is a confounding factor,
o you need to do that in two parts. You need to explain how it affects Variable A and explain how it affects Variable B.
And in those 33 states they had set up what I have here on the right, called the
observed controls.
So it could be the idea that
okay I'm supposed to have an immune response, this now creates an immune response or that kinda thing.
So they're measured
once at the beginning of some time period and then they're measured again multiple times later on. The exact same people are measured more than once. Then those people are compared to themselves. So you have your older measurement and you compare that to your newer measurement.
A cross-sectional study, subjects are measured
once.
• We will come back and talk about longitudinal versus cross-sectional again in module two but wanted t hit it right so you can see the difference. A longitudinal study is
one where a group of subjects is followed overtime
So what happens is we end up in a situation where
our treatment and control groups are not randomly assigned. The treatment is not imposed by a researdcher.
Control group
people that don't get the vaccine
There are people who can't be vaccinated,
people who are allergic to eggs, people who are immune compromised, people who are living someone with chemotherapy, something like htat
Treatment group
people who get the vaccine
Observed controls-So in cases like that we have observed controls, and we also saw that with the polio vaccine a little bit-
people who were given the vaccine compared to just people who weren't. People who have a surgery compared to people who don't. That's a situation where we often do observed controls.
So there's a lot of still research going on about a placebo. But because of our
psychological resposnes and our actual physical go out and play instead of staying holed up and protected, those thing will change how we act and think. And sometimes change the actual chemical composition of what's going on in our bodies.
My notes: So we control for confounding factors beforehand in a controlled experiment by implementing
randomization
• So what they decided to do for the placebo control group and they got 11 states to agree to do the placebo control trial. They took all of the kids in grades 1-3 and sent home the permission forms. Then anyone who didn't consent wasn't considered. That's the no consent group. They did keep track of them, but they didn't include them in the actual comparison. Then they
randomly said, one for you, one not for you, but nobody knew which one was which. This was randomly decided behind the scenes who would get the vaccine and who would get the placebo. And every child received an inoculation
Placebo
receives a fake vaccine so they think they are receiving the vaccine and they go about their life as usual.
o It turned out that Jewish and Muslim men are the ones who traditionally are circumcised. o It also turned out that religion affected the sexual behavior of the women adherence. Specifically that religious Jewish and Muslim women tended to have fewer sexual partners. So it wasn't really the circumcision that was preventing the low cancer rates. The
religion was a confounding factor in that relationship
If an observational study is
replicated over and over and over and over among different people across the world, over long periods of time. Then you can start to say, yes indeed smoking causes lung cancer. And that's what's happned with smoking and lung cancer is that this was replicated.
Observational studies part 1 • WE are going to contrast observational studies with experiments. Observational study definition-
research where the subject's actions determine whether they're in the treatment group or the control group. And that researchers don't come in and determine or influence what the subjects do.
A controlled experiment is one where
researchers determine who is going to be in the control group and who is in the treatment group
• We did a study in America an din Europe and in the south pacific and south America. And all over the world, no matter how you broke it down, people who smoked had higher rates of lung cancer. And over time, it took
several decades, we were able to actually say with confidence that A causes B. But in general you don't want to do that based on even just a couple of observational studies
For almost all healthy humans, vaccines are very safe
side effects are minimal herd immunity protects those who can't be vaccinated no medicine is perfect
Controlled experiments can have
smaller sample sizes, you can usually get good information from just a couple hundred people.
♣ So no matter what you think, maybe you think it affects men and women differently. And I want to have half of the men in treatment and half of the men in control, and half the women in treatment and half of the women in control. And you could say okay I'm going to fix that but anytime a human goes in and says well, I'm gonna put you here and you there and they make those kinds of assignments, inevitably
some kind of bias creeps in. Whereas, if I'm just going to flip a coin each time, then if it lands on heads they go in treatment, if it lands on tails they go in control.
Definition of placebo
something that resembles the treatment that doesn't have the actual live ingredient. It resembles the treatment but lacks the active ingredient.
This is what random means to a statistician-
that I have some outcomes. I know how often I expect them to happen, but when I do it this next particular time, I don't know which one's going to happen.
The problem is
that we don't know and that makes a difference between the treatment and control. So those are the concerns we have
o One problem with the words we have a lot of problem with in statistics as compared to real world is the word random. And especially lately since it's been co-opted by the cool kids. If something weird or unexpected happens, people say, that's random. But that's not what we mean in statistics class. So random doesn't mean weird or haphazard. What it DOES mean to a statistician is
that you have some outcomes that are possible but you don't know which one's going to happen.
• So what they're going to do is take 69 year old veterans of the first World War and they're going to say did you smoke or did you not smoke? • They will look at just that little subgroup between smokers and non-smokers. And if they see a difference in lung cancer weights between those two groups, then
that's a little brick in the wall of smoking causes lung cancer
Accident and vaccine failure are tragic-
the diseases they prevent are much worse. Learn to weight the risks appropriately.
To review-we've talked about randomized, controlled, double-blind experiments. This is
the gold standard for conducting scientific research that we have today. It may not be the best that will ever be, but it's the best that we've come up with so far in being able to tell if A causes B
• Now there were a lot of people yelling and screaming about the safety of vaccines and this was the people in white coats coming to take your children away, lots of hysteria. One nut job who put out that the government had created all these little white coffins because they knew the kids were going to die when they got the vaccine. • The parents also had to consent. They had to convince parents to let their children be part of this trial. It wasn't too hard of a sell because people were afraid. So a lot of people were on board. They took the first and 3rd graders and those who did not consent and called them
the observed controls
Double-blind is where
the participant and the researcher don't know who got the placebo
Blind is where
the participant doesn't know if they received treatment or not
so you're comparing not people, you don't start with one big group and split them in half, but you say okay we've got
this new surgical procedure, let's try it out. We'll compare it to people 10 years ago who got this different surgical procedure and we'll see if it's better.
• Because of how frightening polio was there was a lot of resistance to the idea of a placebo vaccine. In fact Jonas Salk himself felt that it was almost unethical to give children a fake placebo vaccine. • So in the early days (fall of 1953) He had the vaccine and they contacted health departments across 33 different states that have all said, yes okay we want to be a part of this. Utah had a lot to do with the polio vaccines. • Utah played a large role in vaccine and tried a vaccine in Utah county in 1952 (not the Salk vaccine). Takes a large amount of people to
to inoculate the children, to keep track of who gets polio and who doesn't to send those reports back in.
The best way to think about this is
tossing a coin
Cross-sectional studies can tell us about
trends, but they don't generally tell us what happens to people over time
Now in a controlled experiment, confounding factors are controlled for
using randomization
Investigators control for
verb. confounding factor in an observation study. Look at subgroups in the study and compare treatment and control groups.
Placebo
we need a group that thinks they're receiving the vaccine, and then go about their life as usual. You have to think about this, if you know that your child is one of the first or third graders who didn't get the vaccinated you might be more careful about where you let them play. And if you don't know if you're vaccinated or not then they're hoping that they'll be more like the treatment group so that the groups are more similar.
• But it's important to realize that the diseases the vaccines prevent are much worse. And affect many more people than the few accidents we have every year with vaccines. One thing this class will help you do is learn to
weigh these relative risks. There are things we are disproportionately afraid of. Hopefully by getting a little bit more what we call numerate, kind of like literate but with numbers.You'll be able to more accurately judge where those risks lay.
Longitudinal studies are used to help us decide
what happens as people age.
Now the treatment and control gets a little bit fuzzy here as well, because under this way of thinking, the smoking is the treatment and the people who don't smoke is the control. The way you can kind of distinguish this is,
what is the active ingredient? What is the active thing that's different between these two groups? And in this case it would be the smoking
In associations, notice that I've drawn an arrow with an arrow on both sides. We are not able to tell with an association
which causes which.
Use of a placebo/blind/double blind procedures-
yes to placebo. Because they didn't know if they were taking the palmetto or dummy.
There are some common confounding factors and these are the things that
you'll see a lot in surveys and demographic information that is filled out because we have been able to demonstrate that these are pretty common confounding factors
Why would it be important to randomize it?
• All kinds of different bias could creep in. o A doctor who really wanted the saw palmetto to work could maybe try to give it to more healthy looking patients or something like that. So again it's really important to randomize it so that some impartial chance procedure decides who goes in treatment and control
• One more scenario, scenario 3. o This is a convoluted example so we will talk about it in class. Does this study describe an observational study or a controlled experiment?
• Both. It's an observational study between the breast fed children and the formula fed children. But under the formula fed children it's a controlled experiment. • Once they decided I'm going to breast feed or formula feed, then they took the formula people and split them into supplemented formula and nonsupplemented formula. • Let's concentrate on just the breastfeeding versus formula feeding for this study. So if we are just looking at breast feeding versus formual feeding, that makes it an observational study. They didn't say you women breast feed and you women formula feed, that's a personal decision and different factors play into it so they weren't able to do that as a randomized experiment.
♣ What is the treatment group? What is the control group?
• In this case, the treatment group would be, there's some things to consider here. The treatment group could be the formula fed babies or the breast fed babies depending on your point of view of which one is in position of something unnatural or not.
♣ Based on the information you have, is it a longitudinal study or a cross-sectional study?
• Longitudinal because they measured people from birth to 6 years. If there is a passage of time between measurements, that makes it longitudinal
Carefully explain how socio-economic status could be a confounding factor in this research
• See slides-time 5:00 on video. • So here we have, what I want you to do is come up with what is it that this study is associated? • So what are my two things in this study that the research is saying this is what the results are. So breastfeeding and better vision. • So now we want to put socioeconomic status in the box above the two circles. (which is income, your class, upper class, lower class, etc.) So we need to think about how socioeconomic status would affect the choice to breastfeed. Does it affect the choice to breast feed. Are wealthier mothers more or less likely to breastfeed? Are poor mothers more or less likely to breast feed? Is that going to have any kind of impact on that decision?
♣ So while each person was in treatment and in control what was the treatment and control?
• Treatment- swearing • Control- table words
♣ So what we do then is we take subgroups. Let's go back to the smoking example
• You have a whole bunch of people over a long period of time and you kept track of their health information, whether they smoked, how long they smoked, when they quit, did thye get lung cancer or not. • It wouldn't make sense to compare 69 year old veterans of the first world war to a 20 year old woman and say yeah there is a difference there.
Their plan was to take the second graders and give them the vaccine and leave the first and 3rd graders unvaccinated. Then they would
• compare the rates of disease between the 2nd graders and the first and third graders.
Then they take women of the same age and they
• compare then they take people of different SES classes, so wealthy people or college educated people compared to high school educated people. They take office workers compared to steel workers, you know the environment that you're working in in your occupation, and they compare within just the steel workers. Do the people who smoke get cancer more than people who don't smoke. If that's true, then again that's another bvrick in the wall. Another weight on the side of there is a causal relationship here.
And what will happen is about
• half of the men that come through that when you flip the coin they're going to end up in treatment and half of them will end up in control. Same thing with the women, and the smokers, and the different weights, ages, or whatever. About half and half no matter what factor you're trying to control and put in each group. If you randomize you end up with about half and half in each group and this works better and reduces the kind of bias that you might see.
Why are placebos important? There have been a lot of studies about the placebo effect. They will give people fake morphine drips but they feel better after they push the button. There are amazing things that the brain can do that we don't understand, that some people
• just respond to the very idea of having medicine.
• 1st scenario- two studies were done one in 2005 and the other in 2010. Then reported on in 06 and 11. We're going to be reading about the experiment in 2005. (read slide on pg. 3). Identify the treatment and control group-
• treatment group- those who got the palmetto. Control group- those who got the placebo (dummy capsules).
• Scenario 2— o Treatment and control group-
♣ Each individual in the study at one point was in the treatment group and then at another point was in the control group. ♣ What's interesting about this study design is that people were compared to themselves so it doesn't make a whole lot of sense to take a group of people and see how long they can hold their hand in water while they're swearing and then a completely different group who doesn't swear, because there might be some variability, more variability between the two groups. Whereas if you take the same person with their same level of pain tolerance and you let them be in both the treatment and the control group, then you can compare each individual to themselves and you get a lot less variability and a little bit better use of your data.
o Now we're going to do some examples. • Ultrasound and low birth weight o What we wanna think about here is what explains the difference between the observational study and the randomized controlled study?
♣ So what we want to do is start off with the two things that are related. So we read the scenario, they're saying low virthweight is associated with ultrasound exposure. What we want to do is come up with something that might affect why a person would get an ultrasound and also affect why their baby was low birthweight.
Example differences in measurement
♣ The difference in weight as an average. (may make them different) ♣ What kidn of drugs were they taking, what kind of vitamins? o So when she was a kid growing up when she had a headache or her legs hurt, her mom gave her baby aspirin. o Her kids have never had aspirin. In the 80s it started to say don't give aspirin to young children so her children never had aspirin. o So if you're going to compare something that happened to kids in the 80s who maybe had aspiring growing up in the 70s compared to children in the 2000s who were given motrin whenever they had a fever, maybe that makes the difference.
o Use of placebo/blind/ double blind procedures
♣ We could say we don't know why- and sometimes this happens, that people know what they're doing, but tey don't know why they are doing it. That introduces a little bit of blindness to the study. It's better than nothing, but it's not enough to say it was done blind. Obviously the person standing there with their hand in frozen water, knew whether they were saying a swear word or not. So they were able to tell a difference. Maybe you could not show them the clock. And that would kind of blind them, so they weren't maybe counting down and trying to have one or the other be longer. That would be a way to introduce a level of blindness. ♣ But yeah, no placebo in this one. ♣ And limited blindness
Issue with consent
Children with parental consent to participate were different from those without consent. The diseases affected the upper and middle class disproportionately because lower income groups were exposed while still nursing and received protection from that.
If I toss a coin what can happen? Heads or tails. So when you're tossing a coin you can get heads or tails but
I don't know which one it's going to land on. But I can predict pretty well that half of the time it's going to land on heads and half the tiem it will land on tails.
From Slides: Definition of Observational Study
In an observational study, subjects' actions determine whether they are in the "treatment" group or the "control" group. Researchers don't determine or influence what subjects do.
Question- how do you know that neutral words weren't the treatment since it was their hypothesis.
Teacher answer- in this kind of study you could make the argument that the non swearing was the treatment and the swearing was the control. But because they were interested in swearing, that becomes the treatment even though they thought that it would make it worse. That's still considered the treatment group. That's the thing that they're coming in and saying, we want you to do this. They ask the people to say these swear words instead of the neutral words.
Question- is the treatment group always going to be the one that they're trying to prove. Or is it not.
Teacher answer: the treatment group is usually the one where they're trying to impose something on someone, some kind of condition. Like I said, in this one it's a little fuzzy knowing which one's which. Now this is going to come up over and over again this semester, there's not exactly one right answer. We aren't in algebra anymore. The answer is not 5 and no other answer is the answer. There are things to consider